TADAA - Towards Automated Detection of Anaesthetic Activity
0 likes812 views
The document describes the development of a sensor system called TADAA that aims to automatically capture scientific data on anaesthetists' activities. TADAA uses RFID tags and readers to detect anaesthetists' locations and activities. Initial tests show the system can accurately detect 'Recording' and 'Drug Prep' activities, though more work is needed to distinguish all activities. The system was less distracting than human observers. Future work involves refining activity detection, adding more sensors, and developing real-time and retrospective data analysis tools.
TADAA - Towards Automated Detection of Anaesthetic Activity
- 1. HINZ_PostGrad_13 TADAA 1 Aura Laboratory
- 2. HINZ_PostGrad_13 TADAA 2 My road to Health Informatics1 Anaesthetic activity 2 Outline Research methodology3 Design and development 4 Results and Conclusion5
- 3. HINZ_PostGrad_13 TADAA 3 Your Text here Your Text hereYour Text hereYour Text here My road to Health Informatics • 1991 Graduated B.Comm • 91 - 02 Worked in software development • 03 – 04 M.InfoTech at AUT
- 4. HINZ_PostGrad_13 TADAA 4 Your Text here Your Text hereYour Text hereYour Text here Anaesthesia • “Extreme approximation of death” (Euliano, 2004) • “…that even today we understand but partly” (Eger, 2006) • “Every complication has the potential to cause lasting harm to the patient… deviations from the norm must be recognised promptly and managed appropriately” (Aitkenhead, 2007)
- 5. HINZ_PostGrad_13 TADAA 5 Your Text here Your Text hereYour Text hereYour Text here Complications • 49% of preventable adverse events due to „system factors‟ • Poor record keeping • Lack of information • Few standard procedures • Failure to adhere to standards • Poor communication • Organisational culture (Davis, 2003)
- 6. HINZ_PostGrad_13 TADAA 6 Your Text here Your Text hereYour Text hereYour Text here Solutions • Standard procedures • WHO Safer Surgery Checklist • Recording, Adherence to procedures
- 7. HINZ_PostGrad_13 TADAA 7 Task analysis • “A scientific description of the anaesthetist’s task patterns and workload would aid in our understanding of the nature of anaesthetist’s job…and provide a rational basis for making improvements” (Weinger, 1994) • “A scientific description of the anaesthetist’s task patterns and workload would aid in our understanding of the nature of anaesthetist’s job…and provide a rational basis for making improvements” (Weinger, 1994) • Evidence-based medicine requires scientific data to justify improvements
- 8. HINZ_PostGrad_13 TADAA 8 Gold standard for data collection
- 9. HINZ_PostGrad_13 TADAA 9 ‘Scientific description’ ? Observation Detailed ? No Objective ? No Consistent ? Unlikely (Slagle, 2002)
- 10. HINZ_PostGrad_13 TADAA 10 Can we build a system able to capture more scientific data, with less risk of distraction, and lower ongoing cost? Scientific value ? Potential distraction Expensive Automated Observation ?
- 11. HINZ_PostGrad_13 TADAA 11 Design Science methodology (Offermann, 2009) Humans are not ideal instrument for capture of scientific data Anaesthetic record Drug Prep Location + orientation AURA Lab ACSC field test Simulated procedures
- 12. HINZ_PostGrad_13 TADAA 12 Hidden Markov Model Bayesian network A priori rules Body movement Location Object use Voice / sound Video Accelerometer RFID Audio Motion detectors Contact switches Flow meters Sensors Measure Inference Activity detection systems
- 13. HINZ_PostGrad_13 TADAA 13 Rules HMMs (Hidden Markov Models) Proximity LOS (Location + Orientation + Stance) RFID (Radio Frequency Identification) Sensors Measure Inference TADAA
- 14. HINZ_PostGrad_13 TADAA 14 Anaesthetic Record Action Zone Rule: If reader detects any wristband tag then Recording is happening
- 15. HINZ_PostGrad_13 TADAA 15 ARAZ results Lab Field tests Simulations 98 81 100 77 66 96 47 100 0 0 Specificity 97% Sensitivity 69%
- 16. HINZ_PostGrad_13 TADAA 16 Drug Trolley Action Zone • Rule: If reader detects any wristband tag then Drug Prep is happening
- 17. HINZ_PostGrad_13 TADAA 17 DTAZ results Field tests Simulations Specificity 73% Sensitivity 56% 0 100 100 10 64 10099
- 18. HINZ_PostGrad_13 TADAA 18 Activity Fingerprinting • Signal strength „fingerprint‟ built up from multiple tags and readers
- 19. HINZ_PostGrad_13 TADAA 19 Activity Fingerprinting 2 • Fingerprints associated with a location + orientation through SOM clustering
- 20. HINZ_PostGrad_13 TADAA 20 Activity Fingerprinting 3 = Drug Admin IV • Location + orientation sequences associated with activity through HMM analysis 1 second at drug trolley then 2 seconds at machine then 3 seconds at patient
- 21. HINZ_PostGrad_13 TADAA 21 Activity Fingerprinting 4
- 22. HINZ_PostGrad_13 TADAA 22 AF results Lab Field tests Simulations SOM accuracy 99% HMM accuracy 97% SOM accuracy 88% HMM accuracy 10% SOM accuracy 97% On new data 66%
- 23. HINZ_PostGrad_13 TADAA 23 Distraction • Rated on VAS, converted to 0-100 0 10 20 30 40 50 60 70 80 Tags Readers Observer Distraction - Tags & Readers vs Observer (n=20)
- 24. HINZ_PostGrad_13 TADAA 24 TADAA Observer Hardware Readers x3 Tags x16 Cabling Laptop $450 $950 $200 $600 Tablet PC $1000 OTS Software COM monitor $50 Labour Install (4 hours) $100 Ongoing (annual) Replace tags $190 Wage $40000 ? Cost
- 25. HINZ_PostGrad_13 TADAA 25 Conclusion • ARAZ very good at sensing Recording activity • DTAZ good at sensing Drug Prep – But needs more rules to distinguish other activity at drug trolley • AF very good at sensing anaesthetist location + orientation – But requires better activity inference mechanism • RFID sensors less distracting than observers • Higher upfront cost, but lower ongoing cost
- 26. HINZ_PostGrad_13 TADAA 26 Future development • Refine rules – Switching semi-HMM? (Duong, 2005) • Identify lower level activities • Additional sensors – Tag objects - syringes, intubation equipment – Voice detection for „conversing‟ activities – Gaze detection for „observing‟ activities
- 27. HINZ_PostGrad_13 TADAA 27 Future development • Real-time viewer – Communication to staff outside theatre • Repository of activity records – Research unfamiliar procedures – Mine by anaesthetist – Mine by procedure type, patient condition, etc • Formulate „best practice‟ for procedure – Recognise deviations in real-time, raise alarm
- 28. HINZ_PostGrad_13 TADAA 28 Aura Laboratory
- 29. HINZ_PostGrad_13 TADAA 29 References Aitkenhead, A. R., Smith, G., & Rowbotham, D. J. (Eds.). (2007). Textbook of Anaesthesia (Fifth ed.): Elsevier Limited. Davis, P., Lay-Yee, R., Briant, R., Ali, W., Scott, A., & Schug, S. (2003). Adverse events in New Zealand public hospitals II: preventability and clinical context. New Zealand Medical Journal, 116(1183). Duong, T. V., Bui, H. H., Phung, D. Q., & Venkatesh, S. (2005). Activity Detection and Abnormality Detection with the Switching Hidden Semi-Markov Model. Paper presented at the IEEE Conference on Computer Vision and Pattern Recognition. Euliano, T. Y., & Gravenstein, J. S. (2004). Essential Anaesthesia From Science to Practice. Cambridge, UK: Cambridge University Press. Kohonen, T. (2008). Data Management by Self-Organising Maps. Paper presented at the IEEE World Conference on Computational Intelligence, Hong Kong, June 1-6. 29 Anaesthesia > Task Analysis > TADAA > Evaluation > Conclusion
- 30. HINZ_PostGrad_13 TADAA 30 References Peffers, K., Tuunanen, T., Rothenberger, M. A., & Chatterjee, S. (2008). A Design Science Reseach Methdology for Information Systems Research. Journal of Management Information Systems, 24(3), 45-77. Slagle, J., Weinger, M. B., Dinh, M. T. T., Brumer, V. V., & Williams, K. (2002). Assessment of the Intrarater and Interrater Reliability of an Established Clinical Task Analysis Methodology. Anesthesiology, 96(5), 1129-1139. Smith, A. F. (2009). In Search of Excellence in Anesthesiology. Anesthesiology, 110(1), 4-5. Weinger, M. B., Herndon, O. W., Zornow, M. H., Paulus, M. P., Gaba, D. M., & Dallen, L. T. (1994). An Objective Methodology for Task Analysis and Workload Assessment in Anaesthesia Providers. Anesthesiology, 80(1), 77-92. 30 Anaesthesia > Task Analysis > TADAA > Evaluation > Conclusion
Editor's Notes
- #16: Lab and field tests measured response rates when writing on different parts of the clipboard.Lower rates in field test because clipboard was on metal work surface of anaesthetic machine.
- #18: Field tests measured response rates at different areas of the drug trolley.
- #20: Polar plots show 6-point ‘fingerprints’ – signal strength values from 2 tags at 3 readers
- #21: Very simple explanation of how sequences of fingerprints build up an activity
- #23: Field tests highlight existing SOMs much less accurate on new data => inconsistency of signal strength from day-to-day
- #27: Short- to medium-term work